Since the beginning of spoken language understanding research in the 1970s, people have dreamed of communicating with all kinds of devices and appliances using speech recognition. Today, much of this dream has been realized thanks in part to the development of slot-filling techniques for dialogue management. Slot-filling involves using a template or frame for recognizing an utterance where slots represent a particular type of data, such as a date, a stock ticker, etc., which take on certain values, such as Monday, MSFT, etc. When an utterance is recognized in a speech application, a dialogue management component tries to assign the most likely values for the slots of the utterance.